HP Z820, ESXi 6.5d, AMD GPU passthrough issue

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

michrech

New Member
Jul 24, 2017
10
0
1
47
Host : HP Z820, 32GB ECC DDR3, single E5-2670 (have also tested with an E5-2660) - I'm waiting for a second heatsink to come in so I can install the second CPU.

ESXi 6.5.0d is installed. Windows 10 VM is created. When a GPU is added to the VM (I've tried a Radeon HD 7470, HD 6970, and RX480), and then the VM is started, the system hangs -- the console keyboard doesn't do anything, WebUI stops responding, can no longer ping the host. I have to hold the power button to power it off, and then I can power it on normally.

I *can* start up a WHS2011 VM with the onboard LSI controller passed to it, and I can start the same Win10 VM with the onboard USB3 controller attached. It's *only* when I attach one of the afore mentioned GPUs to a VM, then try to power it on, that the host hangs.

None of the devices are installed in PCIe slots three or four (since those are tied to CPU1, which is currently not installed).

Any ideas?
 

MiniKnight

Well-Known Member
Mar 30, 2012
3,072
973
113
NYC
BIOS for vt-d and IOMMU? Didn't someone say there's a vt-d bug with the early E5's? Maybe fixed with microcode so need a new BIOS?
 

michrech

New Member
Jul 24, 2017
10
0
1
47
BIOS for vt-d and IOMMU? Didn't someone say there's a vt-d bug with the early E5's? Maybe fixed with microcode so need a new BIOS?
BIOS has been upgraded from J63 v3.50 to J63 v3.91, which is the latest available. This issue happens with an E5-2660 and an E5-2670.
 

Rand__

Well-Known Member
Mar 6, 2014
6,626
1,767
113
While I did read through that thread, the information in that thread appears to be specific to nVidia, and in that thread the OP is actually able to start his VM without the host hanging. I can't even start the VM -- within about 5 seconds of doing so, the entire host (*all* processes) hangs.
Yes but the basic process works for AMD as well (memory reservation, cpu flags etc)
 

marcoi

Well-Known Member
Apr 6, 2013
1,532
288
83
Gotha Florida
Did you try to run the server with a base install of Win10 without esxi? It might be a hardware issue like no enough power to card which is being missed due to running esxi as the host OS.
I would put a local install of w10 then test to make sure everything works, before continuing to troubleshoot.
 

michrech

New Member
Jul 24, 2017
10
0
1
47
Did you try to run the server with a base install of Win10 without esxi? It might be a hardware issue like no enough power to card which is being missed due to running esxi as the host OS.
I would put a local install of w10 then test to make sure everything works, before continuing to troubleshoot.
Yes -- it worked fine. I'd hope it's not a lack of power issue, considering this system has an 1100+ watt power supply... ;)
 

marcoi

Well-Known Member
Apr 6, 2013
1,532
288
83
Gotha Florida
when you start up the server, do you see the monitor hooked up to the video card showing esxi starting? Then do you see the display hang on start up screen while the system still is booting?
 

Patriot

Moderator
Apr 18, 2011
1,450
789
113
Well that is the SSPEC for the C2 stepping which is the re-release of sandy that supports vt-d.
So this should be doable. SR0H8 = c1 and does not have vt-d functionality.
Have you tried vtd passthrough on kvm or another virtualization platform?

For gpus over 4gb of vram there are esxi specific issues and bios setting that need to be set for memory address support. Look for anything in the bios specifying >4gb addressing and 64bit addressing.... and large bar.
 

michrech

New Member
Jul 24, 2017
10
0
1
47
I was under the understanding it was related to ESXi 6.0+ versions.
If that's the case, I didn't do any of it on my Z800 (the X5677 based system)... I just set the card for pcipassthrough, added it to the VM, reserved all of the RAM, and started up the VM. :)
 

michrech

New Member
Jul 24, 2017
10
0
1
47
when you start up the server, do you see the monitor hooked up to the video card showing esxi starting? Then do you see the display hang on start up screen while the system still is booting?
This is the behavior I see on my Z800 (it has ESXi 6.5 and two video cards). Works great.

On the Z820, the 7470 is the card it uses to boot / show the ESXi console UI. When I start the VM with the RX480 attached to it, the system hangs.

If I start the VM with the video card the system used to boot, the screen blanks (as is normal), then the system hangs.

If I boot the system with a single video card in it (doesn't matter which of the three), attach it to the VM, and start the VM, the screen blanks and the system hangs.
 

marcoi

Well-Known Member
Apr 6, 2013
1,532
288
83
Gotha Florida
did you try cleaning the bios with setting defaults then putting on the specific cpu settings needed?
Also did you try a fresh install of esxi 6.5 and then try 6.0 if 6.5 fails?
Does the z800 have an internal video card, either though cpu or onboard? Does the z820 have one?
 

michrech

New Member
Jul 24, 2017
10
0
1
47
did you try cleaning the bios with setting defaults then putting on the specific cpu settings needed?
Also did you try a fresh install of esxi 6.5 and then try 6.0 if 6.5 fails?
Does the z800 have an internal video card, either though cpu or onboard? Does the z820 have one?
Yes, yes, no, no, no, in that order. ;)

I plan to try 6.0 on it this weekend... ;)
 

seed

New Member
Aug 30, 2017
1
0
1
40
Hi,







The only way i was able to make it work was changing the /etc/vmware/passthru.map



I changed the existing entrance for NVIDIA devices from bridge to link and also add a similar line for my ATI card:



# ATI

1002 6818 link false

1002 aab0 link false

# NVIDIA

10de ffff link false

Hope it help... :)